Yidu Tech's Xu Jiming: Driving Deeper Healthcare Solutions with Large Models

The “2024 World Digital Health Forum,” part of the 2024 Zhongguancun Forum, was held in Beijing. The event was hosted by Tsinghua University and organized by the People’s Government of Haidian District (Beijing), the Chinese Institute of Electronics, the Clinical Medical College of Tsinghua University, and the China Association for Promotion of Industrialization of Science and Technology. The forum brought together 19 academicians, 70 hospital presidents, and more than 100 leading experts, scholars, entrepreneurs, and investors from around the world in the fields of digital health and smart healthcare to discuss the theme “Harnessing Digital Intelligence to Usher in a New Era of Healthcare.” Xu Jiming, CEO and Co-Founder of Yidu Tech, was invited to attend the forum and engage in in-depth discussions with industry experts on the new blueprint for building autonomous and controllable vertical large language models for hospitals.

At the forum, addresses were delivered by Li Bin, Deputy Director of the National Health Commission; Wang Chen, Vice President of the Chinese Academy of Engineering and Academician of the Chinese Academy of Engineering; Ma Jun, Deputy Mayor of Beijing Municipality; Mustafa Shehu, President of the World Federation of Engineering Organizations; Martin Taylor, WHO Representative in China; and Dong Jiahong, Dean of the School of Clinical Medicine at Tsinghua University, Academician of the Chinese Academy of Engineering, and Chairman of the Forum Presidium.

Academicians, heads of research institutions, and renowned experts and entrepreneurs in the digital health sector from China and abroad delivered speeches, offering insights into innovative technologies, development roadmaps, cutting-edge advancements, and the deepening of application scenarios within the global digital health industry, while jointly strategizing for the high-quality development of the digital health cause. Xu Jiming, CEO and Co-founder of Yidu Tech, shared his perspectives on “Building Autonomous and Controllable Vertical Large Language Models for Hospitals,” which drew significant attention from attendees.

Speech by Xu Jiming, CEO and Co-Founder of Yidu Tech

In the medical field, the application of large models can accelerate hospitals’ independent research and development in medical AI, mitigate the risk of technological bottlenecks, and serve as a crucial strategy for digital transformation, enhanced service efficiency and quality, and stimulated scientific innovation. The development of autonomous and controllable domain-specific large models by hospitals is becoming a key driving force behind the intelligent upgrading of the healthcare industry.

Xu Jiming pointed out that the path to building an autonomous and controllable large language model for the medical vertical is not smooth, requiring direct confrontation with several core challenges:

First, in terms of data, building large language models (LLMs) specialized for the medical domain relies heavily on high-quality medical data. The collection, cleaning, and processing of such data not only present high technical barriers but also involve stringent privacy protection and security considerations.

Secondly, the issue of computing power should not be underestimated. Most healthcare institutions struggle to access the robust computational capabilities required to run large models, and this challenge is compounded by high computing costs and the scarcity of efficient computing resources, creating a significant barrier.

Furthermore, challenges at the algorithmic level are manifested by the lack of end-to-end algorithmic capabilities. This is particularly evident in the healthcare sector, a highly specialized field with stringent accuracy requirements, where existing technological standards are still insufficient to fully meet demand.

Finally, the ambiguity and fragmentation of application scenarios pose a significant challenge: not only are the objectives in certain scenarios insufficiently defined, but even in well-defined scenarios, constructing sustainable business models and ensuring reasonable returns on substantial investments remain critical issues.

In the face of challenges, Yidu Tech leverages its decade of experience in medical big data governance and collaborates with numerous hospitals to continuously deepen its understanding of medical scenarios and optimize models. In partnership with Ascend AI, it has developed an integrated solution for large model training and inference. Built upon a foundation model pre-trained on hundreds of billions of refined tokens, this solution offers a comprehensive, out-of-the-box toolchain covering data management, model management, training management, evaluation management, and application management. This not only significantly lowers the barrier to entry for large model development and deployment but also provides efficient local processing and personalized training capabilities, enabling high-quality data to participate in large model training under secure and trustworthy conditions.

An integrated training and inference solution for large models empowers users who understand specific application scenarios to independently build large models and new digital intelligence centers that comply with information technology innovation (Xinchuang) standards and ensure autonomous controllability. Its core advantages include:

● Independent + Secure: Yidu Tech provides a one-stop solution encompassing hardware, models, operating platforms, and applications, supporting users in independently implementing data pre-training and scenario-specific fine-tuning. This enables users to flexibly customize and optimize models according to their specific needs, ensuring open and secure application deployment.

● Professional + Multi-Scenario: The model can undergo enhanced training tailored to hospital settings, with its capabilities continuously strengthened to meet specialized medical demands. Meanwhile, it supports diverse application scenarios, fulfilling the needs of clinical care, education, research, and management.

● No-code development + ease of use: Yidu Tech is committed to lowering technical barriers, enabling users to build applications without writing code. Its out-of-the-box operational platform and visual interface allow users to get started easily and perform efficient data processing and model application.

Furthermore, Yidu Tech has established a mature toolchain that provides end-to-end services spanning data processing, model training, model evaluation, and model deployment. It also offers flexible and autonomous model capabilities, enabling seamless integration with existing in-hospital products and services.

Yidu Cloud, a subsidiary of Yidu Tech, provides hospitals with one-stop solutions by building a “dual middle-platform” architecture, namely the “Data Middle Platform” and the “AI Middle Platform.” The “Data Middle Platform” is responsible for integrating, governing, and securely managing hospitals’ vast data resources, ensuring data quality and compliance. The “AI Middle Platform,” leveraging a robust library of algorithmic models, enables hospitals to rapidly build and train customized models tailored to their specific needs without requiring in-depth understanding of underlying algorithms, thereby achieving “zero-code” operation.

In this way, hospitals can fully leverage their data advantages to build large models tailored to their specific needs, thereby maximizing the utilization of their proprietary data assets and realizing the maximum value of their data.

“Leveraging large language model (LLM) technology to enhance the efficiency of data production and processing enables governed data to drive business scenarios while simultaneously serving as training corpora for LLMs. This creates a virtuous flywheel effect that not only improves data usability but also enhances the applicability of intelligent technologies,” stated Xu Jiming. He added that, built upon this “dual middle-platform” architecture, hospitals can flexibly deploy diverse application scenarios, fully unlocking the new potential of smart healthcare.

Xu Jiming stated that, leveraging large model technology, Yidu Tech has upgraded its existing solutions across multiple scenarios, including data governance, hospital management, clinical research, and clinical diagnosis and treatment.

In clinical diagnosis and treatment, large language models support automated medical record generation, quality control of medical records, and the provision of auxiliary diagnostic and therapeutic recommendations, significantly alleviating the workload of clinicians. At the patient service level, they empower the entire workflow of smart hospital services, enhancing the patient healthcare experience.

At the research level, Yidu Tech’s large language model can rapidly comprehend clinical researchers’ study designs, swiftly generate query criteria, and match patients in seconds, significantly enhancing the efficiency of clinical research and accelerating scientific output.

At the hospital management level, leveraging the intelligence of large language models enables hospitals to conduct refined operations. By asking questions in natural language, the system automatically analyzes data and provides visualized answers, which not only improves decision-making efficiency but also offers managers specific action recommendations.

Xu Jiming stated that Yidu Tech is currently leveraging large language models as its core to drive deeper-level healthcare solutions.

This presentation not only provides an in-depth analysis of Yidu Tech’s exploratory practices in the field of large medical models, but also offers forward-looking insights into the future development of the entire healthcare industry. The industry has good reason to believe that existing hospital data platforms will unlock greater value with the support of AI middle platforms and large models. As a pioneer in medical intelligence, Yidu Tech is actively promoting the construction and development of autonomous and controllable large models, and remains committed to integrating cutting-edge large model technologies with clinical practice to drive the intelligent transformation and high-quality development of the healthcare sector.